A novel deep neural network based marine predator model for effective classification of big data from social Internet of Things

• Published in: Concurrency and computation 2022-11, Vol.34 (25), p.n/a
• Main Authors: Shaji, B., Lal Raja Singh, R., Nisha, K. L.
• Format: Article
• Language: English
• Subjects: Algorithms; Artificial neural networks; Big Data; Classification; deep recurrent neural network; Energy consumption; Internet of Things; Machine learning; marine predator; modified relief; Neural networks; Predators; social internet of things; Water treatment; Water treatment plants
• ISSN: 1532-0626; 1532-0634
• DOI: 10.1002/cpe.7244

Summary Social Internet of Things (SIoT) is considered as one of the most recently evolved topics that connects people and object, object and object as well as people and people. The SIoT and big data provide an exact representation of IoT and social system for human progression characterization. Numerous machine learning techniques are employed to classify the data gathered from SIoT in a more powerful way. In this article, a deep neural network based marine predator (DRNN‐MP) algorithm is proposed in classifying big data. Here, an adaptive Savitzky–Golay filter is employed for selecting the subset and to eliminate undesirable data, as well as different noises. The big data databases are reduced using a Hadoop map reducing framework thereby enhancing the performances of the proposed approach. In addition to this, a modified relief technique is employed to select optimal features thereby performing better classification. The testing and training process based on the proposed approach for optimal classification of features from the big data considers four different databases namely coronary illness, GPS trajectories, localization data, and water treatment plant obtained from UCI machine learning repository. In addition to this, the proposed approach is evaluated for diverse performance measures namely accuracy, precision, specificity, sensitivity, throughput, and energy consumption. Finally, the proposed approach is compared for various metrics to illustrate the effectiveness of the system and the results demonstrated that the accuracy of our work is 98.25%.